As part of a jam-packed day of AMD product news, moments ago AMD’s CEO, Dr. Lisa Su got off the stage, wrapping up her suite of announcements. The highlight of which is AMD’s new family of video cards, the Radeon RX 5700 series. AMD first teased these back at the tail-end of Computex a few weeks ago, and while the cards won’t actually launch until July, AMD has opened the floodgates on information about these cards – pricing, expected performance, architecture – so let’s get to it.

The Radeon RX 5700 series – which I’ll call the 5700 series for short – are AMD’s new family of mid-to-high end video cards. Within AMD’s product stack these cards essentially replace AMD’s previous RX Vega 64/56 parts, offering similar-to-better performance at lower prices, lower power consumption, and with newer features. To be clear, these are not flagship-level video cards, and at some point in time Vega 64/56 will get true successors in the form of faster, more powerful high-end video cards. But within AMD’s product stack and in the broader market, this is where the new cards will land.

These new cards from AMD are part of their first wave of cards based on their new RNDA architecture family. We’ll get into (excruciating) detail about that at a later time, but at a high level RDNA makes some pretty radical shifts in how AMD’s underlying GPU architecture works, more than earning the new name and realigning our performance expectations for AMD video cards. RDNA ultimately seeks to boost both AMD’s workload efficiency – that is, getting more work done with the same resources – as well as their power efficiency, in order to improve their competitiveness in the PC video card market. Pioneered in the Navi family of GPUs, the RDNA architecture will be the basis of AMD products for a long time to come; and not just PC GPUs, but consoles (Xbox and Playstation), mobile, and whatever other deals AMD can land.

But getting back to the matter at hand, AMD is launching two 5700 series cards here next month. At the high end we have the fully enabled Radeon RX 5700 XT (yes, those insufferable suffixes are back), which sports 40 CUs and a peak clockspeed of over 1900MHz. It’s partner in crime will be the suffix-free Radeon RX 5700, which is the traditional second-tier part that cuts back on some functional units and performance in the name of offering a lower-priced card (and letting AMD salvage Navi chips). These parts, AMD tells us, will be competitive with the GeForce RTX 2070 and RTX 2060 respectively, though of course this is something we will determine for ourselves once we have them in for testing.

AMD Radeon RX Series Specification Comparison
  AMD Radeon RX 5700 XT AMD Radeon RX 5700 AMD Radeon RX 590 AMD Radeon RX 570
Stream Processors 2560
(40 CUs)
2304
(36 CUs)
2304
(36 CUs)
2048
(32 CUs)
Texture Units 160 144 144 128
ROPs 64 64 32 32
Base Clock 1605MHz 1465MHz 1469MHz 1168MHz
Game Clock 1755MHz 1625MHz N/A N/A
Boost Clock 1905MHz 1725MHz 1545MHz 1244MHz
Throughput (FP32) 9.75 TFLOPs 7.9 TFLOPs 7.1 TFLOPs 5.1 TFLOPs
Memory Clock 14 Gbps GDDR6 14 Gbps GDDR6 8 Gbps GDDR5 7 Gbps GDDR5
Memory Bus Width 256-bit 256-bit 256-bit 256-bit
VRAM 8GB 8GB 8GB 4GB
Transistor Count 10.3B 10.3B 5.7B 5.7B
Typical Board Power 225W 180W 225W 150W
Manufacturing Process TSMC 7nm TSMC 7nm GloFo/Samsung 12nm GloFo 14nm
Architecture RDNA (1) RDNA (1) GCN 4 GCN 4
GPU Navi 10 Navi 10 Polaris 30 Polaris 10
Launch Date 07/07/2019 07/07/2019 11/15/2018 08/04/2016
Launch Price $449 $379 $279 $179

To start things off, as always we have the specs. It’s best to be clear now that while the raw specifications are helpful in understanding the basics of these cards, due to the RDNA architectural transition the numbers can be deceiving. In particular, RDNA incorporates a number of changes to improve both compute efficiency/utilization and reduce memory bandwidth needs. So if anything, these specifications understate the 5700 series cards by a decent degree, as in practice they’re going to be more efficient per clock than their Polaris predecessors.

And I’m using AMD’s Polaris cards here as my point of comparison – despite the fact that these new cards will perform more like Vega 64/56 – because at a hardware level Navi 10 replaces Polaris 10 as a mid-range(ish) GPU. AMD’s first 7nm GPU for this market, which is being fabbed over at TSMC, measures in at 251mm2, pacing in 10.3 billion transistors into that modestly-sized die. This is a bit larger than the 232mm2 Polaris 10, and incorporates 80% more transistors. So there’s a whole lot more hardware at work here, which for AMD should translate into a good deal more performance.

Diving into the 5700 XT, AMD’s full-fledged Navi card will attempt to put its best foot forward in terms of securing a new spot for AMD in the video card market, and in showing off the RDNA architecture. This is a 40 CU part, with clockspeeds peaking at 1905MHz. New to the Navi generation is a figure AMD is calling their “game clock”, which is analogous to NVIDIA’s turbo clock, and is a conservative estimate of what the average GPU clockspeed is during normal games. To be sure, AMD’s clocking behavior hasn’t really changed – they still try to boost as high as they can, as much as power and thermals allow – but this value is intended to offer better guidance to buyers about what the hardware will typically do.

Looking at these clockspeed values then, in terms of raw throughput the new card is expected to get between 9 TFLOPs and 9.75 TFLOPs of FP32 compute/shading throughput. This is a decent jump over the Polaris cards, but on the surface it doesn’t look like a huge, generational jump, and this is where AMD’s RDNA architecture comes in. AMD has made numerous optimizations to improve their GPU utilization – that is, how well they put those FLOPs to good use – so a teraflop on a 5700 card means more than it does on preceding AMD cards. Overall, AMD says that they’re getting around 25% more work done per clock on the whole in gaming workloads. So raw specs can be deceiving.

Meanwhile on the front and backends respectively of the new GPU, 5700 XT can spit out 4 rendered polygons per clock, and more still when primitive shading is employed. In order to consume all of the pixels that will come flowing out of that process, the GPU ships with 64 ROPs on the backend, twice as many as on AMD’s Polaris cards (or the same number as the Vega cards. We’ll get into architectural matters later, but AMD has put some work in here in order to improve the throughput of these blocks, so the 5700 XT on paper looks like a reasonably well-balanced architecture.

Feeding the beast that is AMD’s Navi 10 GPU are the company’s new GDDR6 memory controllers. While AMD as a company has their arms deep in the development of GDDR memory, for product cadence reasons they are becoming the second company to employ the new memory type,  behind NVIDIA. So as we’ve seen in other products, GDDR6 stands to significantly improve the amount of memory bandwidth AMD has to play with; going from 256GB/sec on comparable Polaris cards to 448GB/sec on these new Navi cards. Compounding this, AMD is now employing delta color compression throughout virtually their entire chip, so memory bandwidth efficiency as a whole is improving. Along these lines, the new GPU employs a new cache heiarchy. The nuts and bolts of this we’ll save for another time, but it ultimately keeps traffic local to the GPU and better avoids using expensive off-die GDDR6 bandwidth when it can be avoided. Overall between its 64 ROPs and significant compute throughput, 5700 XT can eat a lot of bandwidth, and AMD intends to be well-prepared to feed it.

Finally, let’s talk about power consumption. For this generation AMD is sticking with their Board Power figures, which means they’re largely comparable to past AMD cards. In the case of the 5700XT, this is a 225 Watt card, similar to the RX 590 and (on paper) a bit more than the Vega 56. The card draws its power from a combination of the PCIe slot and external power connectors, relying on an 8-pin + 6-pin configuration there. At 225W it’s definitely not a lightweight card when it comes to power consumption, and in our full review we’ll have to see how this translates to real-world performance, and if AMD has tuned this card more towards performance than power efficiency.

Cooling this card will be a largely traditional AMD blower. AMD is employing an aluminum shroud and backplate here (similar to the Vega 64), with a vapor chamber drawing heat up from the GPU to the heatsink. AMD tells us that the blower itself has been further optimized for air flow and noise, and the company has set the default acoustic limit to a relatively low 43dB. Ultimately AMD is trying to strike a new balance between the benefits of the blower design and noise; blowers work far more consistently, which AMD considers desirable for their wide range of customers, but by tuning the blower and capping the noise a bit lower, they’re trying to keep it from being quite so audible in quiet environments.

Radeon RX 5700

Not to be entirely overshadowed, below the Radeon RX 5700 XT we have the vanilla Radeon RX 5700. This card is largely cut from the same cloth as its faster XT sibling, trading off some performance for lower power consumption and lower pricing.

The lesser of the 5700 cards ships with 36 CUs enabled, and more modest clocks. The average game clock is rated for 1625MHz, with a maximum boost of 1725MHz, meaning that for compute/shader/texture workloads it should deliver around 87% of the 5700 XT’s performance. Meanwhile for ROP and geometry throughput, we’re looking at 93% of that performance.

Meanwhile nothing changes for the 5700 relative to the XT card when it comes to memory. It gets the same 8GB of 14Gbps GDDR6. So the 5700 should be even better fed than its full-fledged counterpart, relatively speaking.

The overall drop in performance also means power consumption has come down. The card is rated for a board power of 180W, which is comparable to what the RX 580 was rated for. For anyone crunching the numbers at home, the 45W drop in rated power consumption is even greater than the rated drop in performance, so it’s likely that the 5700 will end up being the more power efficient of the two cards.

Product Positioning & the Competition

Last but certainly not least of course is the details of next month’s launch, and how AMD’s new cards stack up to the competition – both AMD and NVIDIA.

Like the rest of AMD’s new 7nm consumer hardware, the two Radeon cards will be launching on July 7th. The 5700 XT will hit the shelves at $449. Meanwhile it’s smaller sibling, the 5700, will be a $379 card.

Next month’s launch is a traditional, driven-from-the-top full reference card launch. This means AMD’s blower-style reference cards will be what you find on the shelves on the first day. Semi-custom and custom cards are of course in the works, but those will come at a later time. Meanwhile, unlike NVIDIA, AMD isn’t doing a Founders Edition program here, so once custom cards do come out, barring any market changes they should be priced the same as AMD’s reference cards.

Interestingly, even though this is a new GPU family on a new architecture (on a new process), AMD is doing a game bundle of sorts. Bundled with the Radeon RX 5700 series cards is a 3 month subscription for Microsoft’s new Xbox Game Pass program. The company’s new all-you-can-eat game subscription service, despite the name it applies to PC games as well. I would be lying if I said I wasn’t a bit off-put by the idea of including what amounts to a trial subscription as a bundle – as opposed to games you own – but it’s certainly different. And I suspect there was some wheeling & dealing by Microsoft to promote the new service.

Looking at the AMD product stack then, the new Radeon 5700 cards are an interesting addition to AMD’s lineup. They will remain below the Radeon VII as AMD’s fastest card – the Vega derivative is still a tier above – but they are supplanting the Vega 64/56 and then some. According to AMD’s own data, the 5700 XT is on average 14% faster than the Vega 64, and closer to 30% faster than the Vega 56. AMD doesn’t provide similar data for the 5700 (vanilla), but with those numbers the card should easily best the Vega 56.

AMD has been drawing down Vega card inventory for a while now, so if you look at the few cards left on the market, you’ll find that they’re a fair bit cheaper than the new Navi cards. So for the moment they are options as cheaper alternatives, but as is usually the case here, this isn’t a situation that will last (and won’t come with any of Navi’s benefits, obviously). Still, I’m curious to see just how close (or far apart) the two families really end up.

Instead the big competitive question is going to be how all of this compares to NVIDIA’s current-generation GeForce RTX 20 series cards. NVIDIA kicked off that launch almost a year ago, and the unimpressively priced cards have been ruling the roost for a while now.

According to the slides AMD has provided, the $449 5700 XT should beat the $499 RTX 2070 by a few percent. That said, vendor benchmarks must always be taken with a sizable grain of salt, as vendors like to put their products in a good light. Credit to both AMD and NVIDIA here, they’ve actually been pretty decent as of late, so I suspect AMD’s numbers are close to what we’ll find with the hardware. In which case I’m expecting an earnest 2070 competitor, though we’ll see if the 5700 XT can consistently beat it.

A Quick Note on Architecture & Features
Comments Locked

326 Comments

View All Comments

  • Hixbot - Saturday, June 15, 2019 - link

    I'm not following your point. Nvidia's terrible pricing are not being challenged by AMD. If it were, I think they would get a lot of credit from myself and others, just as they are in the CPU market. Nobody is praising Nvidia here, I was just expecting more value from AMD.
  • Korguz - Saturday, June 15, 2019 - link

    Hixbot, i think amd would still be being harped on, and what i mean is.. they would still be criticized like they are now for not having ray tracing, and cause of the price of the cards.. instead.. i think the common complaint would be.. amd had an extra year to work on their cards, and all they can do is match the performance ?? they suck.. and their cards are still over priced...
  • BenSkywalker - Saturday, June 15, 2019 - link

    So what performance level are you seeing in your games with RTX on using your 2070 and where do you think it should be?

    I just played through Quake 2 RTX on a 2060 and I thought it was great. Been playing metro exodus and Tomb Raider with ray tracing on too, not having problems with either of them either. Minecraft isn't my thing but my kids think the ray tracing in that game is great too(that one will run on older hardware).

    Part of the disconnect in this conversation is people talking about how bad performance is at 4k ultra with ray tracing when mid range cards can't run these games art those settings without ray tracing anyway.
  • Korguz - Saturday, June 15, 2019 - link

    BenSkywalker i dont have any 20 series cards.. i have a 1060, but going by the reviews.. and word of mouth with those that do have a 20 series card, it seems the performance isnt there. maybe thats it, resolution 1440 or higher.. but it seems if you mention you play @ 1080p, you get made fun of.
  • BenSkywalker - Sunday, June 16, 2019 - link

    I'll say all the people I know IRL that actually play the games that have tried ray tracing on with RTX hardware have thought performance was solid, on a 2060. Now to be fair no competitive shooters were used online, just actual gamers playing actual games. As an example Tomb Raider max settings without ray tracing at 1440p is close to identical performance as 1080p with ray tracing at ultra settings. Now you can run RT at medium 1440p and be quite playable(mid 40s to mid 50s), but between the two settings, double blind, everyone I had compare said the 1080p with ray tracing was much better(full disclosure, I picked a spot with ray traced shadows on screen).

    But my tournament level frame rates are down..... As opposed to what image enhancing technology?

    Five years from now it'll be a joke that anyone argued against it in the first place, most of those that are, rabidly, will deny they ever did.
  • Korguz - Sunday, June 16, 2019 - link

    BenSkywalker, what games are you trying RT with ?? im not sure.. but tomb raider isnt really all that taxing, is it ? i would assume the " competitive shooters were used online " you mention.. may be a lot more taxing.. and the performance may not be all that bearable with RT on, on a 2060.
    im not arguing against it, but, maybe be unlike some, for the price, it just isnt worth it... yet... i am going to assume you are in the US, but for those of us in canada, ( maybe other parts of the world as well ) take your US prices and add 200 at the low end, to around 500 at the top end and decide of the prices are worth it.. 2060's here start at $500 ( when not on sale ), and go as high as $2100 for a lot of those i know.. it just isnt worth it, as we have bills to pay, kids to feed etc...
  • BenSkywalker - Monday, June 17, 2019 - link

    Metro exodus, Tomb Raider, Minecraft and Quake 2 RTX. Take Metro Exodus, fps are higher using 100% shader resolution with ray tracing on then ray tracing off with shader resolution maxed out. Digital foundry made a video about Quake 2 RTX, really gives a great example of the impact(performance drop is *huge*, so is the visual impact).

    So you say it isn't worth the premium, in no way whatsoever would I say that's wrong, I know first hand readily available disposable income isn't always sitting around in large quantities, but what about when there is no premium?

    That's what I'm seeing with this launch. The 5700 is more expensive than the 2060, barely edges it in traditional rendering and doesn't give the option to play with ray tracing *at all*.

    There's only a handful of games and it's a big performance hit, both completely valid, but the pricing issue is kind of out the window now that AMD has decided to avoid the value position.
  • Korguz - Monday, June 17, 2019 - link

    BenSkywalker, looking that req's for Metro exodus, im a little surprised that it runs that well for you :-) but to then add in a 22??? year old game, that would run ( exaggerating here ) 400 fps on modern hardware, add RT to it, and have it run at " only " 200 fps, is a little moot... what would the performance be, and i dont mean patching in RT support, but if the game was made for RT from the start ? the friends i talked to.. and i think this is part of the reason why that dont think it is worth it to get an 20 series card, yet, is be cause with Metro exodus, Tomb Raider, Minecraft and Quake 2 RTX, they dont play any of those games.. so the money spent for RT and the other features the 20 series brings to the table, wouldn't be used. to compare the 5700 series to the low end 20 series.. is also a little lopsided, as amd is aiming these 2 cards well above that, so comparing the 2060 series to say a 5600 or even a 5500 type series might be a little more even. but it really comes down to how much is one willing to spend on a video card, how much they can afford to, and would the games one plays, see any benefit from that purchase? one of the friends i talked to.. is one of those with a lot of disposable income, and even he says the 20 series isnt worth the cash right now :-)
  • BenSkywalker - Monday, June 17, 2019 - link

    The canned bench from metro exodus is like a torture test, the game runs quite nicely. Quake 2 is actually 32 years old, you were way off on you're estimates(think more like 1.5k to sub 100), but I'd say watch the digital foundry video.

    Now, the rest of your comments, they are directly refuting AMD's claims. AMD is calling the 5700 a 2060 competitor, saying it is 10% faster for 8.5% more money. That's not my interpretation, that's what they have come out and said. AMD is saying, based on their hands picked benches that they are going to charge almost exactly the same $/fps as nVidia but no option for ray tracing. Again, this isn't me spinning anything, it's all in their slides, their words.

    Is it unreasonable to assume AMD chose benches that made them look slightly better than average? Combine that with the price premium and you may find that real world has a factory overclock 2060 at the same price as the 5700 MSRP is dead even with AMD, but with the option of playing with RTX.
  • BenSkywalker - Monday, June 17, 2019 - link

    That should say Quake2 is 22 years old.

Log in

Don't have an account? Sign up now